Search papers, labs, and topics across Lattice.
D vision tasks, with representative examples including diff-pc [21] and Mvdream [27]. DifFUSER[14] first applied diffusion to multimodal fusion by generating BEV features to handle noise; MDD [10] leveraged diffusion for multimodal integration; and CoGMP [4] incorporated generative map priors via diffusion to reduce transmission overhead. Building on these ideas, this work injects diffusion-driven denoising objectives directly into the feature space, enabling unified modeling of diverse and compound noise sources for multi-agent feature reconstruction and cooperative fusion. 3 Method The goal of cooperative perception is to enable a group of NN connected agents 𝒜={a1,…,aN}\mathcal{A}=\{a_{1},\dots,a_{N}\}, to achieve comprehensive scene understanding by sharing information. The system processes collective data 𝒳={Xj}j=
1
0
3
4
Achieve superior cooperative perception by adaptively compensating for temporal latency and noise, outperforming existing methods in complex traffic scenarios.
